Ellipsoid

Deep Breakdown

What it tells me (Situation Awareness): An ellipsoid is a 3D surface defined by $\frac{x^2}{a^2} + \frac{y^2}{b^2} + \frac{z^2}{c^2} = 1$, where $a$, $b$, $c$ are the semi-axes along the $x$, $y$, $z$ axes respectively. It is the 3D generalization of an ellipse. The volume formula $V = \frac{4}{3}\pi abc$ generalizes the sphere volume $V = \frac{4}{3}\pi R^3$ (setting $a = b = c = R$). In DA/DS: the ellipsoid is the confidence region of a multivariate Gaussian distribution. The set $\{x : (x-\mu)^T \Sigma^{-1}(x-\mu) \leq c^2\}$ is an ellipsoid with axes aligned to the eigenvectors of $\Sigma$ and axis lengths proportional to $\sqrt{\lambda_i} \cdot c$. This is central to Mahalanobis distance, PCA, Gaussian mixture models, and covariance visualization.

Flowchart

Real-world use

Where it is used

Visualizing covariance structure of multivariate Gaussian distributions (PCA confidence ellipsoid), Mahalanobis distance thresholds (the boundary $D_M = c$ is an ellipsoid), Gaussian mixture model cluster shapes, 3D bounding volumes in computer graphics and robotics (minimum volume enclosing ellipsoid), and quality control (tolerance ellipsoid for multivariate measurements).

Problem it solves

In 3D, spheres are too restrictive for representing data spread: data may be stretched along certain directions. An ellipsoid captures directional variance. Given a covariance matrix $\Sigma$, the 1-sigma ellipsoid shows exactly which 3D region contains 68% of the distribution's probability mass.

How it solves the problem

The equation $\frac{x^2}{a^2} + \frac{y^2}{b^2} + \frac{z^2}{c^2} = 1$ is the boundary where each axis contribution is normalized by its semi-axis length. Points strictly inside satisfy $< 1$. Volume $V = \frac{4}{3}\pi abc$ comes from integrating the sphere formula with each axis rescaled: if you stretch a unit sphere by $a$, $b$, $c$ along each axis, the volume scales by $a \cdot b \cdot c$.

Example in a business

A DS fits a 3-feature Gaussian anomaly detector on normal sensor readings. The alert boundary is an ellipsoid in 3D feature space with axes aligned to the principal components of $\Sigma$ and lengths proportional to $\sqrt{\lambda_i \cdot \chi^2_{0.99}}$. The DS visualizes this ellipsoid to validate that the detector is capturing the right region of normal behavior.

The formula, piece by piece

• $a, b, c$: semi-axes of the ellipsoid along $x$, $y$, $z$ (all $> 0$)
• $\frac{4}{3}\pi$: the same constant as in the sphere volume formula
• $abc$: product of the three semi-axes; replaces $R^3$ in the sphere formula
• Surface equation: $\frac{x^2}{a^2} + \frac{y^2}{b^2} + \frac{z^2}{c^2} = 1$
• Interior: $\frac{x^2}{a^2} + \frac{y^2}{b^2} + \frac{z^2}{c^2} \leq 1$
• Special case $a=b=c=R$: $V = \frac{4}{3}\pi R^3$ (sphere)
• Special case $a=b \neq c$: oblate ($c < a$) or prolate ($c > a$) spheroid
• Surface area: no closed form in general; requires elliptic integrals

The formula in words

"Scale a unit sphere by $a$ along $x$, by $b$ along $y$, and by $c$ along $z$. The volume of this stretched sphere is the sphere volume $\frac{4}{3}\pi$ times the product of the scaling factors $abc$. The boundary is the set of points where the sum of normalized squared coordinates equals 1." Tiny example: $a=3$, $b=2$, $c=1$. $V = \frac{4}{3}\pi(3)(2)(1) = 8\pi \approx 25.13$. Compare sphere $R=3$: $V = \frac{4}{3}\pi(27) = 36\pi \approx 113.1$. The ellipsoid has much less volume because it is flattened.

Critical Concepts

• Mahalanobis ellipsoid: the boundary $D_M = c$ is an ellipsoid with axes = eigenvectors of $\Sigma$ and lengths $= c\sqrt{\lambda_i}$.
• PCA confidence ellipsoid: for a $d$-dimensional Gaussian, the 95% region is the ellipsoid with $D_M^2 \leq \chi^2_{0.95, d}$.
• No closed-form surface area: unlike volume, ellipsoid surface area requires elliptic integrals. Use scipy.integrate or approximations in practice.
• Minimum volume enclosing ellipsoid (MVEE): the smallest ellipsoid containing all points. Used in robust statistics and optimal experiment design.
• Degeneracy: if any semi-axis is 0, the ellipsoid degenerates to a lower-dimensional shape.
• Parametric form: $(x,y,z) = (a\sin\phi\cos\theta, b\sin\phi\sin\theta, c\cos\phi)$ for $\phi \in [0,\pi]$, $\theta \in [0,2\pi)$.

Step-by-step calculation (tiny example)

Step 1: Set up the ellipsoid Semi-axes: $a=4$, $b=3$, $c=2$. Equation: $\frac{x^2}{16} + \frac{y^2}{9} + \frac{z^2}{4} = 1$.

Step 2: Compute volume $V = \frac{4}{3}\pi(4)(3)(2) = \frac{4}{3}\pi(24) = 32\pi \approx 100.53$.

Step 3: Check if a point is inside $p = (2, 1, 1)$: $\frac{4}{16} + \frac{1}{9} + \frac{1}{4} = 0.25 + 0.11 + 0.25 = 0.61 < 1$. Inside. $\checkmark$

Code examples

Beginner example (Python)

import numpy as np

def ellipsoid_volume(a: float, b: float, c: float) -> float:
    """Volume of ellipsoid with semi-axes a, b, c."""
    return (4.0 / 3.0) * np.pi * a * b * c

def is_inside_ellipsoid(point: np.ndarray, a: float, b: float, c: float) -> bool:
    """Check if a 3D point is inside/on the ellipsoid."""
    x, y, z = point
    return float(x**2/a**2 + y**2/b**2 + z**2/c**2) <= 1.0


a, b, c = 4.0, 3.0, 2.0
print(f"Semi-axes: a={a}, b={b}, c={c}")
print(f"Volume = (4/3)*pi*a*b*c = {ellipsoid_volume(a, b, c):.4f}")
test_points = [
    np.array([2.0, 1.0, 1.0]),
    np.array([4.0, 0.0, 0.0]),  # on boundary (x-axis endpoint)
    np.array([3.0, 2.0, 1.5]),
]
for pt in test_points:
    val = pt[0]**2/a**2 + pt[1]**2/b**2 + pt[2]**2/c**2
    inside = is_inside_ellipsoid(pt, a, b, c)
    print(f" {pt}: normalized_sum = {val:.4f} -> {'inside' if inside else 'outside/boundary'}")

Beginner example (R)

# -- Ellipsoid volume and point classification ---
ellipsoid_volume <- function(a, b, c) (4 / 3) * pi * a * b * c
is_inside_ellipsoid <- function(point, a, b, c) {
  (point[1]^2 / a^2 + point[2]^2 / b^2 + point[3]^2 / c^2) <= 1
}
a <- 4; b <- 3; c <- 2
cat(sprintf("Volume = %.4f\n", ellipsoid_volume(a, b, c)))
test_pts <- list(c(2, 1, 1), c(4, 0, 0), c(3, 2, 1.5))
for (pt in test_pts) {
  val <- pt[1]^2/a^2 + pt[2]^2/b^2 + pt[3]^2/c^2
  status <- ifelse(is_inside_ellipsoid(pt, a, b, c), "inside", "outside")
  cat(sprintf(" (%.1f, %.1f, %.1f): normalized_sum=%.4f -> %s\n", pt[1], pt[2], pt[3], val, status))
}

Professional example (Python)

import numpy as np
import pandas as pd
from scipy.stats import chi2
np.random.seed(42)

# -- Business scenario: 3D Gaussian confidence ellipsoid for sensor anomaly detection ---
N_TRAIN = 300
D = 3

# Simulate correlated 3D normal data
Sigma_true = np.array([[4.0, 1.5, 0.5], [1.5, 2.0, 0.3], [0.5, 0.3, 1.0]])
mu_true = np.array([2.0, 1.0, 0.5])
L = np.linalg.cholesky(Sigma_true)
X_train = np.random.randn(N_TRAIN, D) @ L.T + mu_true

# Estimate mean and covariance
mu_hat = X_train.mean(axis=0)
Sigma_hat = np.cov(X_train.T)

# Eigendecomposition: ellipsoid axes aligned to eigenvectors
eigenvalues, eigenvectors = np.linalg.eigh(Sigma_hat)  # sorted ascending

# Chi-squared threshold for 99% at D=3
chi2_threshold = chi2.ppf(0.99, df=D)
semi_axes = np.sqrt(eigenvalues * chi2_threshold)

# Volume of the 99% confidence ellipsoid
volume_99pct = (4.0 / 3.0) * np.pi * np.prod(semi_axes)
results = pd.DataFrame({
    "component": [f"PC{i+1}" for i in range(D)],
    "eigenvalue": eigenvalues.round(4),
    "semi_axis": semi_axes.round(4),
})
print(f"3D Gaussian confidence ellipsoid (99%, chi2_threshold={chi2_threshold:.2f}):")
print(results.to_string(index=False))
print(f"\nEllipsoid volume (99% confidence region): {volume_99pct:.4f}")
print(f"Sphere volume with same max semi-axis: {(4/3)*np.pi*semi_axes.max()**3:.4f}")
print(f"(Ellipsoid is smaller: it accounts for covariance shape)")

Source: domain knowledge

#linear-algebra#spatial-distance-norms-and-analytic-geometry#analytic-geometry-shapes-perimeter-area-volume